Model profile
Deepseek V3.2
DeepSeekdeveloper
2025-12-01release date
#79 / 267overall rank
23eval lineages
Evidence summary
Deepseek V3.2 has an estimated overall rank of #79; its 90% source-sensitivity interval is #46–#151. Its behavior-only rank is #68; company governance moves the combined estimate to #79. Published evidence spans 23 evals and 7 of 7 behavior components. Its strongest relative result is TAC (base_welfare_rate, #13 of 68); its weakest is CAIS Risk Index (agent_red_teaming, #42 of 43).
Compare this model
Only models sharing at least one published sub-eval are listed.
Official and reference links
- Artificial Analysis ↗deepseek-v3-2
- Hugging Face ↗deepseek-ai/DeepSeek-V3.2
- OpenRouter ↗deepseek/deepseek-v3.2
- Model card ↗Exact model document · DeepSeek · first party
Published eval results
Rank is within that sub-eval. Black marks the observed result; the grey dot marks the value implied by the global rank. Values stay on each source’s native scale.
| Eval / sub-eval | Rank | Value | Distribution | Better | Source |
|---|---|---|---|---|---|
| AA-Omnisciencehallucination_rate | #190 / 311 | 0.8435 | ↓ lower | Source ↗official | |
| AgentAbstainabstain | #12 / 17 | 52.1 | ↑ higher | Source ↗official | |
| AgentAbstaincar | #13 / 17 | 50.2 | ↑ higher | Source ↗official | |
| AgentAbstainpaired | #12 / 17 | 41.4 | ↑ higher | Source ↗official | |
| Alignment Leaderboardcorrigibility | #6 / 24 | 4.476 | ↑ higher | Source ↗official | |
| Alignment Leaderboardhonesty | #11 / 24 | 3.613 | ↑ higher | Source ↗official | |
| Alignment Leaderboardnon_manipulation | #9 / 24 | 3.557 | ↑ higher | Source ↗official | |
| Alignment Leaderboardrobustness | #12 / 24 | 3.453 | ↑ higher | Source ↗official | |
| Alignment Leaderboardsafety | #8 / 24 | 4.019 | ↑ higher | Source ↗official | |
| Alignment Leaderboardscheming | #7 / 24 | 4.006 | ↑ higher | Source ↗official | |
| ANIMAscore | #6 / 19 | 0.7244 | ↑ higher | Source ↗official | |
| BullshitBench v2clear_pushback_rate | #83 / 105 | 0.115 | ↑ higher | Source ↗official | |
| CAIS Risk Indexagent_red_teaming | #42 / 43 | 93 | ↓ lower | Source ↗official | |
| CAIS Risk Indexbioweapons_assistance | #43 / 48 | 94 | ↓ lower | Source ↗official | |
| CAIS Risk Indexhle_overconfidence | #30 / 49 | 58.8 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmachiavelli | #27 / 45 | 89.6 | ↓ lower | Source ↗official | |
| CAIS Risk Indexmask | #42 / 51 | 53 | ↓ lower | Source ↗official | |
| CAIS Risk Indextextquests_harm | #38 / 48 | 21.6 | ↓ lower | Source ↗official | |
| Cisco AI Defense Rolling Single-Turn Leaderboardsingle_turn_attack_success_rate | #98 / 105 | 76.72 | ↓ lower | Source ↗official | |
| DystopiaBenchbasaglia_score | #39 / 50 | 69.37 | ↓ lower | Source ↗official | |
| DystopiaBenchbaudrillard_score | #42 / 50 | 70.33 | ↓ lower | Source ↗official | |
| DystopiaBenchhuxley_score | #35 / 50 | 75 | ↓ lower | Source ↗official | |
| DystopiaBenchlaguardia_score | #37 / 50 | 69.03 | ↓ lower | Source ↗official | |
| DystopiaBenchorwell_score | #32 / 50 | 72.83 | ↓ lower | Source ↗official | |
| DystopiaBenchpetrov_score | #38 / 50 | 77.8 | ↓ lower | Source ↗official | |
| HUMAINE Trust, Ethics and Safetytrust_ethics_safety_score | #13 / 54 | 28.02 | ↑ higher | Source ↗official | |
| LiveSecBenchethics | #9 / 43 | 75.93 | ↑ higher | Source ↗official | |
| LiveSecBenchfactuality | #26 / 43 | 40.76 | ↑ higher | Source ↗official | |
| LiveSecBenchlegality | #15 / 43 | 65.16 | ↑ higher | Source ↗official | |
| LiveSecBenchprivacy | #26 / 43 | 37.7 | ↑ higher | Source ↗official | |
| LiveSecBenchpsychological_health | #14 / 43 | 61.46 | ↑ higher | Source ↗official | |
| MACHIAVELLIdeception_relative_random_pct | #27 / 50 | 89.6 | ↓ lower | Source ↗official | |
| Manager Coercion Benchcoercion_ladder_depth | #27 / 31 | 9 | ↓ lower | Source ↗self run | |
| MORUscore | #4 / 13 | 79.14 | ↑ higher | Source ↗official | |
| RefusalBenchyouden_j | #17 / 19 | 0.01333 | ↑ higher | Source ↗official | |
| SABERoverall_safety_rate | #4 / 13 | 27.6 | ↑ higher | Source ↗official | |
| SABERscenario_a_safety_rate | #8 / 13 | 27.31 | ↑ higher | Source ↗official | |
| SABERscenario_b_safety_rate | #3 / 13 | 36.09 | ↑ higher | Source ↗official | |
| SABERscenario_c_safety_rate | #3 / 13 | 21.67 | ↑ higher | Source ↗official | |
| SM-Benchadversarial | #62 / 73 | 75.61 | ↑ higher | Source ↗official | |
| SM-Benchambiguous_interpretation | #64 / 73 | 70.54 | ↑ higher | Source ↗official | |
| SM-Benchanti_hallucination | #39 / 73 | 91.1 | ↑ higher | Source ↗official | |
| SM-Bencheq_boundaries | #27 / 73 | 67.42 | ↑ higher | Source ↗official | |
| SM-Benchoverfit | #59 / 73 | 44.81 | ↑ higher | Source ↗official | |
| TACbase_welfare_rate | #13 / 68 | 36.54 | ↑ higher | Source ↗official | |
| TukaBenchafri_jbb_cultural_asr | #3 / 6 | 19.2 | ↓ lower | Source ↗official | |
| TukaBenchafri_jbb_harm_asr | #4 / 6 | 16.9 | ↓ lower | Source ↗official | |
| TukaBenchafrijail_mono_asr | #3 / 6 | 20.1 | ↓ lower | Source ↗official | |
| Vigil Mental Health Safetyoverall_score | #15 / 23 | 35 | ↑ higher | Source ↗official |
Values evaluations
Descriptive values and political-framing results are separate from safety/ethics ranks. Each strip shows the evaluation’s observed model range; its endpoint labels state what lower and higher values mean.
UGI Political Values
Agent-ValueBench Moral Foundations (MFT08)
Agent-ValueBench HEXACO
| Dimension | Value | Distribution |
|---|---|---|
| Openness to experience | 6.8 | |
| Honesty-humility | 5.4 | |
| Extraversion | 6.4 | |
| Agreeableness | 6.6 | |
| Conscientiousness | 7.6 |